A SOTA World Model for Risk Anticipation

BADAS 2.0
The world's best incident prediction model, period.

BADAS 2.0 is a world model that has internalized physics and causality. Wetrained it on 2M real-world driving clips. We fine-tuned NVIDIA's Cosmos on thesame data. BADAS wins – at 91x fewer parameters. We publish everything. Try ityourself.

99.4% Average Precision
91% Early Warning
4.6% False Alarm Rate
Cloud to Edge
01 – Live Demo

Run BADAS 2.0 on Your Own Video

Watch the world's best incident prediction and understanding model in action.

Upload Your Own Video. See What BADAS 2.0 Sees.

Drag an EGO POV video. Get predictions, attention heatmaps, and natural language reasoning – in seconds.

Choose Video File
Supported: MP4, MOV, AVI – up to 60 seconds
Real Incident 1
Real Incident 2
Real Incident 3
Real anonymized edge device incidents – BADAS 2.0 prediction overlay on actual near-collision events from Nexar's 350K camera fleet
Bumper Cars
Formula 1
BADAS 2.0 collision probability overlay -- probability ramps from near-zero as danger develops.
Bumper Cars
Formula 1
Generational leap - BADAS 2.0 detects danger earlier with higher confidence and 58% fewer false positives
Bumper Cars
Formula 1
BADAS 2.0 vs Cosmos-BADAS - purpose-built vision transformer outperforms fine-tuned VLM on temporal discrimination
Bumper Cars
Formula 1
Attention heatmap - visualizes where the model focuses as danger develops
Bumper Cars
Formula 1
BADAS-Reason - natural language explanations describe what the model sees and why it predicts danger"
02 – The Open Challenge

0.994 AP vs 0.940. Same Data. Different Architecture.

We fine-tuned NVIDIA's Cosmos-Reason2-2B on the exact same 2M training clips. BADAS 2.0 achieves 99.4% AP. Cosmos achieves 94.0%. At 91x fewer parameters, our smallest model still beats their largest. We publish our benchmarks and invite direct comparison.

Metric
BADAS 2.0
Cosmos Fine-tuned
Average Precision
99.4%
94.0%
Early Warning Recall
91.3%
48.3%
Architecture
V-JEPA2 (Attention)
Autoregressive
Smallest Model
22M params
2B params (cloud only)
Training Data
2M real-world clips
2M real-world clips (same)
Explainability
Native attention maps
None
The Efficiency Gap: Fine-tuning improved Cosmos by +18.5 pp AP – but at 2B parameters, it still sits 5.4 pp below BADAS 2.0 (94.0% vs 99.4%). BADAS 2.0 Flash Lite (22M) outperforms Cosmos Fine-tuned (2,000M) by +4.4 pp AP while being 91x smaller.
Early Warning Recall
48.3%
Cosmos Fine-tuned
91.3%
BADAS 2.0
In collision anticipation, late is too late.
The Efficiency Gap
Flash Lite (22M) outperforms Cosmos Fine-tuned (2,000M) – 91x smaller
BADAS 2.0 — 300M, 99.4% AP
BADAS 1.0 — 300M, 96.0% AP
Flash — 86M, 99.0% AP
Flash Lite — 22M, 98.4% AP
Cosmos Fine-tuned — 2B, 94.0% AP
Cosmos Vanilla — 2B, 75.6% AP
The Crisis

The Work Zone Crisis

898
Fatalities/year in U.S. work zones (FHWA)
$1B+
Unpaid roadway enforcement fines, from NYC alone
482M hrs
Lost annually to work zone congestion (FHWA)

Every year, U.S. work zones claim nearly 900 lives, strip cities of over a billion dollars in unpaid fines, and add half a billion hours of congestion to American roads. The root cause is always the same: no one knows what’s actually happening out there.

NYC DOT processes 700,000 street construction permits a year, but a State Comptroller audit found nearly 300,000 received no inspection at all. 150 field inspectors. 6,000+ miles of streets. Every un-inspected permit is a compliance gap, a safety risk, and an uncollected fee hiding in plain sight

Powered by AI. Driven by 350,000+ dashcams.

Real-time road intelligence. Updated daily by the vehicles already driving your streets.

CityStream Work Zones uses Nexar’s nationwide fleet of 350,000+ AI-powered dashcams to continuously detect, classify, and monitor road work zones — at city scale, at state scale, across the entire U.S. main road network.

No survey trucks. No drones. No field visits required. Ground-truth intelligence, delivered continuously.

4 steps

How It Works

Dashcams Collect

350,000+ AI-powered dashcams already on U.S. roads passively capture ground-level imagery as they drive through work zones. No special hardware deployment needed.

AI Detects

Nexar’s AI analyzes imagery to identify and classify 8 types of work zone elements — from grabber cones to arrow boards to barriers. Each detection is GPS-tagged, time-stamped, and accompanied by photographic evidence.

Permit Cross-Reference

Detected work zones are cross-referenced against permit databases in real time. Overstays, early starts, and unpermitted activity are flagged automatically — with photo evidence attached.

Map & Monitor

Detections are aggregated, geolocated, and displayed on an interactive map. DOTs can monitor all work zones across their jurisdiction, track changes over time, and export via API or WZDx-compliant feed.

Benefits

Why CityStream

See Every Work Zone Without Leaving Your Desk

DOT staff currently send crews to physically drive through work zones to verify conditions. CityStream replaces manual compliance checks with continuous remote monitoring — dramatically reducing officer time on site visits and enabling proactive enforcement without adding inspection burden.

Know What’s Really Happening,Not What’s on the Permit

Detected work zones are automatically cross-referenced against permit data. The system flags overstays, early starts, and unpermitted activity — with photographic evidence. Ground truth, not paperwork.

The Freshest Road Datain the Industry

Highways refreshed in days. Arterials in about a month — roughly 50x faster than traditional survey cycles. 98% of U.S. main roads visited monthly. This isn’t once-a-year survey data. It’s continuous monitoringat scale.

WZDx Compliance, Out of the Box

CityStream outputs WZDx-compliant data natively — feeding accurate work zone data to mapping services, connected vehicles, 511 systems, and other agencies. 23+ state agencies already participate. Stay ahead of the federal standard with zero extra lift.

Revenue Recovery at Scale

Revenue Recovery at ScaleWork zone permit violations are a massive, largely untapped revenue source. NYC alone has $1.4B in uncollected enforcement fines. CityStream’s compliance layer identifies non-compliant and unpermitted work zones — opening a path to recover fees that currently go undetected. In the City of LA, the platform is funded through recovered compliance and violation fees. It pays for itself.

Reduce Congestion. Improve Traveler Information

Work zones cause 24% of all non-recurring U.S. traffic congestion. Greater permit compliance accountability reduces unnecessary lane closures. Real-time work zone updates feed directly into 511 systems and WZDx feeds — improving traveler information accuracy across the board.

Our Unfair Advantage

No survey vehicles. No new infrastructure. Immediate insights.

Everything in CityStream is powered by the Nexar dashcam network — 350,000+ AI-powered cameras already on U.S. roads, capturing the world as it actually is

350,000+

AI dashcams deployed nationwide across consumer and fleet customers.

98%

of U.S. main roads visited every month.

80+ million

detections per day.

Privacy-first by design:

on-device anonymization, no route sharing, GDPR and CCPA compliant.

Nexar owns the hardware

end-to-end control over quality, AI processing, and privacy

Mean Time Between Visits - All Road Types

Trusted by Leading DOTs

Deployed across cities and states that can’t afford to guess

City of Los Angeles

3+ years | 23,000 lane-miles

A trusted partner for work zone compliance - LA has relied on CityStream for 3+ years to monitor thousands of active work zones across 23,000 lane-miles."

Partner DOT

1,800+ deployed dashcams

Statewide DOT deployment demonstrating Nexar’s ability to operate at scale across an entire state’s road network.

Company logo
Company logo
Company logo
Company logo
Company logo
Company logo
Company logo
For buyers who know what WZDx is

WZDx-compliant. Out of the box.

The Work Zone Data Exchange (WZDx) is the federal standard for sharing work zone information with vehicles, navigation systems, and other agencies. 23+ states already participate. CityStream Work Zones outputs WZDx-compliant feeds natively — no integration work, no custom build. As connected and autonomous vehicles become the norm, WZDx compliance will be critical infrastructure. You’re already ready.

Privacy by Design

No driver data. No vehicle tracking.
No exceptions.

Nexar’s on-device AI automatically blurs faces and license plates before any image ever leaves the dashcam. Driver routes are never shared. Start and end points are stripped. All tracking IDs are ephemeral — no individual driver can be identified or traced. Zero personally identifiable information is present in any work zone data delivered to customers.

GDPR compliant
CCPA compliant
Privacy architecture reviewed by design, not as an afterthought

See your roads clearly

Schedule a live demo and see CityStream Work Zones monitoring your jurisdiction in real time.

02 – Why the Old Model Breaks

Your Teams Can't Be Everywhere. 
Now They Don't Have To Be.

Work zones are dynamic. Traditional oversight models are not.

By the time an issue is identified through an inspection, report, or complaint, theoperational moment may already be gone.

Field Inspections

Important, but not continuous.

Self-Reporting

Useful, but incomplete.

Complaints & Incident Reports

Reactive, not real-time.
Agencies need better ground truth between inspections, not just more paperwork
after the fact.
03 – The Solution

Know Your Roads as They Are, Not as They Were.

Nexar provides AI-powered road intelligence that detects work zone activity from alarge, continuously refreshed connected camera network covering the majority of U.S.roads.

Using machine learning models, Nexar identifies work zone elements such as cones,barriers, barricades, and signage, then maps them into structured, geolocated workzone records.

This is not just what was planned. It is visibility into what appears to be happening onthe road.

How It Works

Observe

Nexar’s connected cameranetwork captures real-worldroadway conditions at scale.

Detect

AI models identify work zonesignals including cones,barriers, signs, barricades, andarrow/message boards.

Map

Signals from multipleobservations are aggregatedinto mapped work zones withoperational context.

Bring into Workflow

Available through IBM watsonxOrchestrate, this intelligencecan support agency workflowsfor monitoring, review, andresponse.
04 – Use Cases

See Every Work Zone. Act on What Matters.

Improve Visibility into Active Work Zones

See where work zone activity is appearing across thenet work with better ground truth.

Surface Potential Permit Compliance Issues

Identify roadway conditions that may warrant review against agency expectations and permit conditions.

Reduce Blind Spots in Oversight

Support teams with better visibility between inspection sand across fast-changing corridors.

Focus Manual Effort Where It Matters Most

Use roadway intelligence to prioritize field attention more effectively.

Support Operational Workflows

Bring work zone intelligence into agency processes through IBM watsonx Orchestrate.
05 – IBM watsonx Orchestrate

Intelligence That Shows Up Where Your Team Already Works.

Detection alone is not enough. Operational teams need a way to review, route, and acton roadway intelligence efficiently.

Available through IBM watsonx Orchestrate, Nexar’s work zone intelligence can bebrought into agency workflows to support faster, better-informed decisions.

956
Work zone fatalities (FHWA 2022)
$15,870
Per violation, per work zone
50%
Increase in deaths over the past decade
$1.4B
Uncollected fines (NYC alone)

Access road intelligence in a workflow environment

Reduce reliance on disconnected monitoring tools

Support more efficient review and response processes

Bring observed roadway signals closer to operational action

06 – Proof

Backed by the Largest Road Observation Network in the U.S.

350K+
Connected
Cameras
98%
U.S. Road
Coverage
80M+
Objects
Detected Daily
8 in 10
Segments
Refreshed
Weekly
6
Object Classes
Detected
Real-time road observation across the U.S.
This is road intelligence built for real operational conditions, not occasional snapshots.
Nexar road intelligence is trusted by
GM
Waymo
Waabi
Lyft
NVIDIA
TomTom
City of LA
07 – FAQ

Frequently Asked Questions

What does the solution detect?
Cones, barriers, barricades, signage, and arrow/message boards -- aggregated into mapped work zones with geolocation, timestamps, and visual context.
Does this replace inspections?
No. It improves visibility and supports oversight workflows, helping agencies focus manual effort where it matters most.
Does this automatically determine permit violations?
No. It surfaces roadway intelligence and potential issues for review. It does not make compliance determinations.
Why is IBM watsonx Orchestrate part of this?
It provides the workflow environment through which agencies can access and operationalize Nexar's work zone intelligence within their existing processes.
What kinds of agencies is this for?
State DOTs, municipal transportation departments, public works organizations, and other public-sector teams responsible for roadway operations and work zone oversight.
Is this available now?
Yes - available as an off-the-shelf capability. Contact Nexar for deployment details and access through IBM watsonx Orchestrate.
08 – Get Started

Take Control of Every Mile You Manage.

Talk with Nexar about work zone intelligence for transportation agencies and how it can be brought into workflow through IBM watsonx Orchestrate.